METHODS OF ESTI!MATING L-_ .:butIn IUlirwitd2 - CiteSeerX

8 downloads 0 Views 7MB Size Report
T'han an analyst...alI considerations of Soviet ..... follow the procedure used by Rosch and Mervis (1975) to classify vehicies (see also Tversky, 1977: 338).
METHODS OF ESTI!MATING STRATEGIC INTENTIONS

Frank J. Stech Kenneth C. Hoffman Research Report May 1982 Prepared for: Office of Naval Research Department of the Navy Contract No. N00014-81-C-0508

C-"

L-_

v•

LA-

,Bethesda,

Prepared by: MATHTECH, Inc..4630 Montgomery Avenue Maryland 20014 Telephone: (301) 657-1610

.:butIn IUlirwitd2

-0?Y

"

- -- -- - -- -- -

CONTENTS

Page I ,S UM MARY. . . . . . . . . . . . . . . . . . . . . . . . . . .

li.

. . . . . . . . . . . . . . . . .

DESCRIPTION OF THE INTENTION EST!MATION PROCESS

AND

RELATED ANALYTICAL AIDS ........

.......

................

A. Model of Int~entions ...........................................

5

B.

Content of Analytic Methods ................................... C. Analyt"ical v4athods ............................................

17 49

III.

COMPARISON AND EVALUATION OF ANALYTICAL AIDS ........................

IV.

CATALOGUE OF AIDS AND PROCESS SUMMARY............................... 115

v.

BIBLIOGRAPHY .......................

..

. . . ...............

. ...... 138

3on--

'"'cL

NSPEC TED.

t

52

--

SECURITY CLASSIFICATION OF

Block 20.

MI~qS PAOIrWhm Data £nlaftd)

Abstract (continued)

1. A description of the estimation process and of related problems including the outline of a conceptual representation of the process against which specific methods may be evaluated. 2. A description of analytical aids useful in the intention estimation process including Judgmental methods as well as analytic?' aids using techniques of extrapolation, structural, and causal analysis. 3. A comparison of analytical aids in the context of the conceptual estimation process. 4. The evaluation of specific analyti'~al aids. 5. Development of a catalogue of 18 analytical aids matched by methodology to steps in the conceptual estimation process.

.U

:

Unclassified

UNCLASS IFIED SECURITY CLASSIFICATION OF 7)4

PAGE /Whefn Dats Ev~tt.td

REPORT DOCUMENTATION PAGE ON?

4.

TITLE (an

UMBE

j~OVT

READ 24STRUCTIONS

Subtitle)

METHODS OF ESTIMATING STRATEGIC INTENTIONS 7.

BEFORE COMPLETL4G FORIM

CCESION~o

S. TYPE Of REPORT b PERIOO COVERLU

TcnclRpr

6.

AUTWOR.)

CONTRcnial OReporNt USR

41. PERFORMING ORG. REPORT NUMBER

N01-1C00

Frank J. Stech Kenneth C. Hoffman S.

*

______________________________

10. PROGRAM ELCMENT. PROJECT. TASK & WORK UNIT NUMBERS

PERFORMING ORGANIZATION NAME AND ADDRESS MATHECHInc.AREA

4630 Montgomery Avenue

______________

Bethesda, Maryland 20014 II.

t2.

CONTIROLLING OFFIC9 NAME AND ADDRESS

Office of Naval Research, code 430B StretI1. Nrth

800

REIPORT OATII

May 1982

uinc

NMBERC

OF PAGES

2217143 Vrgina Arligto, IT MONITORING AGENCY NAMEL 4 AOORESS(I liflo

tef~amtfr Contro@Wani Office)

IS. SECURITY CLASS. (WItid* e£paO)

154OCL ASSi FI C ATION/ DOWN GRADING SCHEDULE 64. OISTI1IU?IOk STATEMENT (*I Al. Repo"w)

Unl imitedi Aproe fo, pubL@ ;m,%smw, 17.

OSTRIOUTION STATEMENT (oif tAc 66b.4U

oetled ln J16a* ID. It differnt lIma Reps$)

B9. SUJPPLEMENTARY NOTES

Is.

x

UP WOROS (Contim"e

on

tovere aide HInu..ewp ed id"at**p by Weeck rninaot)

Methods of Analysis Military Intelligence Military Psychology Naval Intelligence

Capabilities & Intentions Cognitive Processes Intelligence Estimates Intention Estimation 2.ABSTRACT (Cusaus -

jf

#*areide of nobw.M .emi

fdWtf 104& 1 ("eck suo.)

SThe estimation of strategic intentions isan important and evolving part of the intelligence process. The research program described in this report was undertaken to catalogue a numbcr. of quantitative and behavioral r.aethods of analysis used insupport of variouis phases of the estimation process. Thus, thi! research builds on earlier work performed by MATHTECH inwhich the intention estimation process w~s defined and various supporting methods identified. Specific tasks undertaken inthis research project are: DO

FOI~m.

1473

EDITION OF I NOYV641

OBSOLETE

Ucasfe SECURITY CLASSIFICATION OF ThIS WAGE (When~ Dots, Entered)

I.

SUM4MARY

The estimation of strategic Intentions Is an important and evolving part of

the intelligence process. Soviet naval

A:

The motivation for focusing on evtimates of peacetime

intentions In previous research studies performed by MATHTECH

arises from the Increased contacts between U.S. or Western fleets and the Soviet Navy.

These more frequent and extensive contacts

sea lanes,

in off-shore defenisive zones,

and in proximity to Third World countries may occur in crisis situa-

tions or in situations with the potential for escalation Into a major crisis because of the behavior and responses among the parties Involved.

The way In

which the navies of the West and the Soviet Union conduct themselves in these peacetime and crisis encounters has a considerable Impact on subsequent escalation and responses.

The previous research performed by MATHTECH dealt with an

assessment of methods for estimating peacetime Soviet naval intentions.

In the

courie of that research a number of quantitative and behavioral methods were identified (Stech, 1981).

A follow-on research program, the results of which

are described in this report,

was then undertaken to catalogue those methods In

relation to estimation problems and issues.

I

In addition criteria were estab-

lIshed to guIde the selection of approprIate methods to deal wIth specIfIc estimation

Issues.

As background for this research,

a quite detailed descrip-

tion of the Intention estimation process was developed. The Individual tasks that were performed In this research project are:

TASK 1.

Descr;otlon of the Estimation Process and of Problems that Arise In the Process,

The conceputal model developed !n this task to provide the context for Intention estimates Includes 8 sequential steps;

o o o o o o o o TASK 2:

preceptions (reactions to the environment and situation based on capabilities and behavior), estimates and assessments of current and future capabilities, risks, actions and reactions, Integration and judgements related to evaluation of alternate hypotheses, definition of Intentions based on logic embodied in previous steps, formulation of a strategic concept outlining courses of action. strategy statement to guide future action, development of plan to guide actions, and execut'on of the plan.

Oescrjition of Analytical Aids:

Analytical aids useful In the intention estimation process vary considerably in scope, methodology, and logic. Consistent with the complexity and range of possibilities of Intentions. these aids are similarly complex and wide ranging; there are no individual methods that are to be preferred even in a given class of analytical problems. The methods vary considerably In their dependence on data, judgements, and relational or behavioral descriptions, as well as in their dependence on quantitative techniques. They range from purely subjective or judgemental methods to highly quantitative probabtilstic methods. A general categorization of analytical aids is as follows: o o o o

Judgemental Methods; supported by attributes of sagacity, control, and acumen. Analytical Aids; quantitative methods that support specific elements of the estimation process. Extrapolation Models; based on past events and data, and Structural models; taking Into account structural realities and changes In leadership, physical capabilities, and other Influencing factors.

The third and fourth categories make extensive use of quantitative methods, In one case Involving trend extrapolation, In the other attempting to represent the physical and behavioral relationships that affect Intentions In a given time and place. Clearly, these categories are not mutually exclusive and Intention estimates will be based on some combination of methods drawn from these categories. The art of estimation is to select and blend appropriate methods and the objective of this research project Is to assist in that process through the cataloguing and organization of analytical aids used to support and extend human judgement and Interpretation. The specific analytical aids covered In this review are listed az follows In relation to the elements of Inten4 :-, ,'teaitng:

1.

V A '1

Perceiving Data: categories.

statistical sampling,

record of events, coding

2.

Weighting Data: polIcy capturIng, Bayesian methods, and variance analysis.

3.

Characterizing Data:

4.

Assessing Covariations: bootstrapping.

5.

Cause and Effect Assessment: causae search, causal analysis, search trees, stepping analysts, hypothesis, regression analysis.

6.

Predictions:

7.

Theories:

TASK 3:

memory aids,

Backcast!ng,

correlation

fuzzy sets, factor analysis.

actuarial

models,

backcasting.

boot strapping, decision analysis.

scenarios, Judgement heuristics. etc.

Comparison of Aids and Problems:

This task compares the analytical aids in the context of the Intention estimation process. General criteria that apply to the comparison Include:

o

o o o o o

o o o o

o o

TASK 4:

scope and content of method In relation to estimation problem. (can method Incorporate Information on capabilities, force structure. deployment, logistics, manpower, risk propensity, strategic patterns, responsive patterns. etc). descriptive power inferential power treatment of constraints (operational. tactical, and strategic), logic structure (deterministic. probabilistic, Inference. etc). transparency of relationships and assumptions. analytical method (state of development and prior experience). application and documentation of method, validation, verification, aikd peer review of method. cost of Initial system and upkeep, treatment of risk and uncertainty, and sensitivity (robustness of result and etfect of changes In assumptions).

Evaluation of Aids:

Problems vith particular analytical aids and an assessment of strength and weaknesses are discussed In the context of the Intention estimation process with reference to historical experiences.

TASK 5:

Catalogue of Aids and Progcess SumraCv

:

A summary table relating specific analytical ails. analytical problems, and criteria was developed and Is Included in this report,

3 VT

Chapter II of th I

report covers Tasks I and 2.

the descrIpt l.n of the

intention estimation process and the description of analy t ical aids pertinent to that process. analytical

Chapter

III covers Tasks 3 and portions of 4.

aids and their evaluations.

the comparlson of

The summary catalogue of aids Is pre-

sented In Chapter IV. Major conclusions that are drawn from this project are: 1.

Estimation of intentions may usefully draw upon a variety of analytical aids.

2.

The selection of specific analytical aids and their application to deal with estimation problems depends upon such factors as scope of problem, data availability, and timing. This process cannot be described In cook-book fashion with any absolute guarantees of success. While the process depends a great deal on creativity, criteria have been developed and can prove quite useful as a checklist for the selection and use of aids.

3.

The criteria and evaluation methods employed In this project to catalogue and evaluate aids in the estimation of peacetime Soviet naval Intentions are sufficiently general to apply to a wide range of intelligence, logistic, and strategic analytical aids of Interest to the office of Naval Research.

4

II.

DESCRIPT•ON OF THE INTENTION ESTIMATION PROCESS AND ANALYTIC AIDS

The estimation of Intentions by Intelligence analys-ts can be characterized in a variety of ways and in terms of: the estimation steos taken by the analysts, the analysts' logic and reasoning processes, and the types of evidence and Information analysts might consider to form an Intel I Igence estimate of Intentions.

This chapter uses these three outlooks on Intention estimation to

describe the estimation process and to Identify problems that might affect various steps of the process.

A.

kJOEL OF INTENTIONS This model

.comparison

is intended to provide the context or background for the

and evaluation of analytical aids.

complex definitions,

or simple ones; e.g.,

Intentions can be given very

Lawrence (1972: 83) writes that

Intentions are "desirous foreknowledge or expectant desire."

Basicaily

Intentions and psychological states that involve Mental Images of future events In which the In-tender pictures himself as a participant and makes choices as to which Image he will try to bring to reality. (Griffin 1976: 5).

i•.

An Intelligence professIonal (GazIt, 1980) has divided Intentions estimation ito Identifying a de4i3lon already taken, analyzing responses taken In reae+Ion to actions by others, situation.

and analyzing the outcome of a developing,

BahavIorally,

;rintentionality manifests several

ongoing

features:

expeCtations of the outcome of an act, selection among alternatives for the fulfIllment of a goal,

sustained effort In a given direction In the sequence of

actions taken to make the r'esultIng state resemble the expected state, and ,.

I•-5

flexible plan for using alternative means and actions to compensate and correct for discrepancies between expectations and results (Foster and Brandes, 1980: 326).

Longer,

(Anscombe,

philosophical

1969; Lawrence,

dissertations on the meaning of "IntentionsO

1972; Stech, 1979, ch. 2) are interesting but not

essential to outlining a descriptive model of the elements of Intentions which an Intelligence analyst may need to analyze. in the following model we are assuming that the adversiry, or enemy, or ally, that Is, the party whose intentions the intelligence analyst hopes to ascertain,

Is not psychopathological.

Model

of intentions can be created for

analyzing the psychopathologic adversary (e.g., Langer's, 1972, model of the "mind of Adoif Hitler" created during World War II to aid AlI led Into[Iigenca estimations),

but each such model

Our objective

is

is a special case.

to outlIne a model

that can hope to deal

estimation of the intentions of most normal actors, osychotic disturbances.

with the

functioning wi houl obvious

We do md assume that the adversary is wholly rational,

in the logical, mathemiutical,

or economic meaning of "rational Ity."

In fact,

one I Imitation of the "rational actor model" (as used, e.g., by Allison, 1971, to assess the Cuban Missile CrIsIs)

is

rational 8•e± when they do mathematics,

that a2 human actors are strictly symbol Ic

ogiC,

or economic utility

calculus, and even .hen biases, errors, and paradoxes can lead to behavior that is not complete!y ravional. actor,

The only real requirement for the model is that the

whose behavior the analyst is attemptlng to predict through the analysis

of intentions, not be -Idely perceived as totallyIrrational,

I.e., the modelI

doe" not work If the adversary Is crazy. The qualification analysts' standpoint.

In the preceding paragraph Is fairly impotant from the The customers of

those which hope to estimate Intentions,

6

ntell I Igence estmates.

particularly

will often justify their ignoring sucti

estimates with the or,

issertion that Prime Minister So-and-so

is "unpredictable"

in the extreme case, President Such-and-such is simply "crazy".

jargument

Neither

Is tenable; the first asserts that we are as wIse as we will ever be,

and that the fact 'if it

Is a fact) that no accurate predictions were made In

4 he pa~rt me•ns none wi I

ever be made In the future.

fact +:tat truly crrzy p

lopie are quite easy to predict,

The second Ignores the use very circumscribed

roulines of behavior, and s-iffe. from the lack of behavioral flexibili ty and choice, i.e., their behavlor is compel led, compulsive, and henc,• predictable. The arguments may justify scepticism

In the predictions,

but not complete

ignorance of them. ThIs model (FIgure 1) has two main sources.

its general characteristics

come from a model used In World War II to predict Axis Intentions through The analysis ot propaganda, (1959).

and It has been subsequently systematized by George

Grafted onto this basic skeleton are concepts taken from the general

areas of declsion-making and problem-solving behavior. Is

That Is,

the adversary

perceived to be solving a problem: what actions are needed to reach some

goal; and making a decision: what action should be Implemented to reach the goal.

The analyst's task is to try to penetrate the adversary's problem-solving

and decision-making processes and procedures, I.*.,

to recreate the steps the

adversary has taken to solve the problems and make a decision. The top line of the figure shows Intentions, which can only be Inferred, *3 Intervening Weteen the real world sItuot;on and the adverswy's actions, both of which can be observed dlrect ly, as well as inferr.~1 fros other evicence. The dotted arrow reflects the fact that 2ome of the adversary's actions are not intentlonal but Instead are accidental or •lsL.Jrriages of intenaed actions. Unintended actlon is diffIcult to soparat, analytlcaily from Intended action; There is a tendency to attrlbute al

oI

action to soOe Motive or IntentIcn of the

~-4).

U)

c

9

w E-4

Ul,6

4.1e 10~4-4 >4

0 -4

0D

PC*4

4

0

44 Q

u Q

C

-

0-

0-40

W

".

u~

0 4o

-4

±

.J-

--4

-4

>,6. /LTV-

0

1 om

0

0 _1

E 0

-4d

0-

04 u

0

actor (see Stech,

1980: 104-120 for a discussion of this tendency In naval

analysis). The second

line

of

the figure

reflects

the approach

taken

Intentions, that Intentional behavior reflects problem formulation, making, and then planning for the execution of the decision taken.

here

to

decisionThese steps

are conditioned by the adversary's perceptions of the real world situation (i.e.,

'1

presumed to Influence the actions taken.

I A

the adversary considers reality In formulating the problem) and are

The

sequential

estimation,

steps

of

problem

formulation

(logic

cf

estimation,

and decision/planning have been the subjects of a variety of

theories and concepts used by political scientists to assess governmental actions. third

".-!

Some of these political science formulations are reflected on the

line of the figure.

For example,

generalizations -bout the typical

perceptions and estimations made by a nation faced with various situations contribute to understanding how the nation formulates problems, and offers information on the important dimensions and value criteria the nation will tend to use to structure Its decisions.

Similarly the operational code of a national

elite provides information on the instrumentalities and basic strategic values of the top decision-makers.

Again,

such Information contributes

to

understanding how a real world problem might be solved, or more basically, what will be perceived as a problem In the first place. At the bottom of the figure are seven steps which reflect the process that Intervenes between the stimulus of the real world situation and the 4c;'ion taken in response.

These are represented

in Figure 2 as sequential

steps indicating

activities preparatory to and subsequent to the definition of Intentions. eighth step, action, has been added to Figure 2 for completeness. real world produces an Immediate perceptual

Iq

response.

An

First the

The adversary reacts to

00

r. 0 0

r

1 0

1

4. 0

0 41

0

O )

a1d00

-4.

00 In Q

0

I. m

c

IV

0

.

0

c o

o.'..0 Z

0

0

0 In u

1

U

00

40 000

4

>

0

U

>0 m 10 0

V

00

0

0

0

0

0 0

0

0

0

~

00 0

~00

O0UC

0

4

>.

c

0

0

00 Z0>

-

C 0

0

10

the perceIved envIrcnment specIfIcally by noting needs,

reactIng to Its own

previous actioors (e.g., a previous failure to satisfy a need mIght provide the stImulus for subsequent action),

the actions of opponents to block or facilitate

need satisfaction, the capabilities of the adversary and of opponents to satisfy needs through actions.

In this first step the adversary Is selecting aspects of

the environment to attend (e.g.,

what is the need,

those aspects along important dimensions,

what is the problem),

coding

and generally characterizing the

nature of the environment Into those parts that require action or attention (i.e., Identifying needs), and those that do not. The second step of the Intention process Involves the organization of perceptions, estimates, and expectations

into orderly categories of Information.

Among these categories are the following:

estimates of current and future re-

sources, both for one's own sIde and for the opposite sIde, estimates of possible actions on each side and the risks attached to them and also the utilities attached to the outcome of the various actions.

The latter Information requires

the analyst to form some impression of the adversary's value system.

The ana-

lyst needs to estimate how the adversary evaluated outcomes for each side.

It

Is helpful to separate the elements of this step Into those that deal with Issues of fact (which can also be thought of as predictions) and those that deal with Issues of value and evaluation.

For example,

ting five different modes of attack.

Several

the enemy may be contempla-

Issues of fact must be estimated:

what resources are needed for each mode, what Is the likelihood of each mode succeeding,

what Is the opponent capable of doing In reaction to each mode and

what are the probabilities of each of those reactions.

Additionally,

various

Issues of value must be assessed: how willing Is the adversary to expend resources at various levels (i.e., what Is the cost of the resources),

how good

are the various possible outcomes of the different attack modes (i.e.,

what are

FU

44

________

the benefits) and what are the values to the parties on both sides of the enemy's reactions to the various possible models.

To make these assessments the

analyst must form some impression of how the adversary makes estimates of facts and probabilities and how he assesses values,

both for the adversary's own ac-

tions and for his estimates of the analysts' country's reactions and initlatives. The third step In the intention process Integrates the information on facts and values to determine what actions would be better than other actions and the likelihood of various outcomes and benefits. according to value, risk, probability, to the adversary

cos-,

or any other criteria of importance

decision-making system.

formulating expectations

Possible goals ire then ranked

In this step,

and clarifying those "mental

the adversary

Is

Images of future events"

which might be brought to reality. To assess this step In the intention process the analyst may have to consler the adversary's decision-making rules. of attack offers the adversary tactical advantage, gains,

I.e.,

For example,

suppose one mode

low risks of loss but little

in the way of

low gains.

Another mode of attack offers spectacular

but at the possible risk of huge losses.

If the adversary allows these

dimensions of risk and gain to compensate for each other (i.e.,

a lot of gain

makes up for a lot of risk) the riskier, more advantageous course might receive the higher ranking. trade-offs (e.g., is

However,

If the adversary cannot affort such compensating

because resour.es are so limited that any major risk of loss

Intolerable) then an unacceptabla state on a single dimension,

may be

sufficient to elimin:te that option

Obviously,

from further

such as risk,

consideration.

a similar conclusion might be reached by the analyst who attends to

whether the adversary Is general

y a gambler or conservative.

The decision

analysis approach has the added advantage that It attempts to identify y

12

the

adversary might be conservative or Intrepid. One aspect of the adversary which may play an important role at this step Is doctrine, those prescriptions as to what types of action outcomes should be highly valued (e.g., sleze the high ground) or avoided (e.g., never concentrate in range of enermy guns).

Doctrine can be viewed as codified decisions, or

stereotyped integrations of estimates and assessments from past experience. Such codes may offer Important clues as to how the adversary will Integrate the estimates and assessments bearing on the current problem however,

It is always possible,

that the adversary will flout doctrine, eIther Intentionally (e.g., as

an explicit means to achieve surprise), ignorance or confusion). specific detailed case,

or unintentionally

(e.g.,

out of

It is also possible for doctrine, when appiled to a to yield Incompatible recommendations

(e.g.,

the

principles of war include both "concentration" and "economy of force," two recommendations that are rarely simultaneously compatible),

and thus offering

little guidance to either the adversary decision-maker or the Intention estimating analyst. In the fourth step the adversary selects a subset of goals from the rankIngs created in the third step. adversary's polIcy.

This constitutes the Intention decision, or the

The adversary selects a desired and possible outcome (or

objective) based on the benefits that would seemingly result from It and the apparent costs of the path to that outcome,

given the resources that are

available for accomplishing that outcome, the feasible actions that are required, and the anticipated contingencies of enemy response. Ing this one Involve estimating these aspects of the decision.

The steps precedThe steps subse-

quent to this one entail refinement of the policy selection made In this step. If the challenge to the analyst in the previous step was to narrow down the range of possible courses of action to those the adversary would consider

13

feasible and worthwhile,

the challenge in this step is to determine which

objective or policy the adversary will successful

value most highly.

An analysis of

intention estimation by intelligence analysts in the past (Stech,

1979) reflects two characteristics that might aid other analysts. Successful

Intention estimators used rather explicit analytical models of

how the adversary made decisions.

These models related the general war objec-

tives of the adversary (the two historical cases took place In World War II) the adversary's resources and constraints. by a process of i nf erence).

That is,

to

These explicit models were refined

the analysts determined how evidence of

different kinds would lead them to change the models.

The analysts then sought

the necessary evidence to determine which direction the models should evolve. In effect,

the analyst anticipated the possible behaviors of the adversary and

considered the implication of each possible move for the evolving model adversary's Intentions. to refine the model,

As the evidence became available,

making It more explicit.

Eventually,

of the

the analyst was able the enemy's actions

and reactions could be estimated with considerable accuracy.

This process by

the analysts was one of "sagacity," i.e., understanding how the adversary had behaved

In the past and the nature of his motives and "acumen,"

I.e.,

the

analyst's ability to understand and anticipate the enemy's response and to duplicate the mental

decision-making

ch. 2, on sagacity and acumen).

logic of the adversary

(see Stech,

1981,

To a lesser degree the analysts were able to

use "control" to assist in the development of their models.

That Is,

the

analysts were at times able to control the information or the situation the adversary faced (e.g., by sending false Information through double agents or deception operations,

or by knowing in advance what one's own forces were plan-

ning to do) and could thus use this control to settle open questions about the enemy's reactions and decisions.

14

The fIfth step in the intention process consists of translating the Intention decision,

or polIcy,

into a strategic concept with a general objective and This outline forms the basis for subsequent

general courses of action outlined. detailed planning.

It IdentlfIes whe± the adversary intends to achieve and how

this will be accomplished to a degree sufficiently detailed so that planners can turn this strategic concept into concrete actions. The sixth and seventh steps Involve detailed planning so that the Intention can be carried out In action.

First (step six) planners must determine from the

strategic concept what specific actions and subobjectives are required to attain the overall strategic objective.

The necessary steps must be ordered Into a

sequence of operations so that each sub-step is completed before subsequent steps that build on it are undertaken.

Orders for required resources are issued.

losses calculated.

given to troop and naval sions.

Enemy actions have to be anticipated and

untis.

Training operations focus on anticipated, mis-

Studies are made of uncertain aspects of the plan.

the target area Is Increased.

Warnings are

Reconnaissance of

Deception operations may be undertaken to conceal

these detailed preparations. In the last step of the intention process,

the adversary writes and Issues

the detailed plans for the Intended operation.

This may have the traditional

form of the m IItary or naval operational order,

or It may be far more Informal

and Instantaneous,

for example,

It

may simply consist of the codeword that

signals the execution of an operation that was planned long before.

In this

step, those actors that must carry out the adversary's intentions must carry out the final steps before actual

:1tying forces,

i

subobjectives,

action Itself: assigning responsibilities, specl-

allocating resources,

fueling vehicles,

sending out scouts,

15

issuing maps and plans, etc.

arming

The adversary has not com-

mitted himself to the plan,

but Is ready to do so and has taken all but the

final step of action itself. The description of these seven steps in the intention process makes it clear that the adversary's behavior becomes easier to monitor as the intention process approaches the action step.

More physical activities and preparations

are needed to translate the strategic goal into strategic technique and then into tactical plans then are needed at earlier stages. adversary comes to actually carrying out his Intentions,

Thus the closer the the greater the amount

of physical Intel I Igence the adversary will be forced to generate and, wishes to conceal

his

intentions,

to hide or disguise.

disguising are also actions which generate physical David Kahn (1978: 39-41) Intelligence" and "verbal

makes a useful

Intelligence."

if he

(Sut hiding and

intelligence.)

distinction between "physical

Physical

Intelligence is

der*Ved from

the natural resources, physical installations, numbers of weapons and troops available, volume of commercial trade, and so forth. derived from words through plans, orders, estimates,

Verbal Intelligence Is

morale, perceptions,

intentions,

promises, or motives.

Plans and intentions take time to translate Into physical realIty.

Knowing

an adversary's intentions gives time to react, whereas knowing what actions the adversary has already begun to take may leave Ilittle or no time for reaction. While the latter steps of the

Intention

process

generate

more physical

Intelligence than the earlier steps, the earlier steps may tend to generate more verbal

Intelligence,

mates,

makes judgments and declsons, end so on.

I.e.,

process of the adversary, physical

Information about how the adversary

perceives,

esti-

In modelIng the Intention

the analyst will have to use verbal Intelligence and

Intelligence to Infer the events between the real world situation and

the adversary's actions.

16

B.

CONTENT OF ANALYTICAL METHODS Intentions with certain characteristics are easily predicted.

To the

degree that *he analyst can determine that the intentions are of the more predictable type he can put greater confidence In his estimates. Stable Intentions are predictable. *

Two types of stability apply to Inten-

tions.

Some behaviors are habitual,

drives.

The need of a nation's people for food and shelter during a crisis such

I.e.,

prompted by recurring needs and

as war may lead to an intention by the nation's leadership to support minimal living standards.

Other behaviors are customary or respond to strong cultural

norms and are thus stable so long as these norms remain effective. customary behaviors are highly predictable,

Habitual and

If the analyst develops the abl ity

to recognize the repetitive pattern and the stimuli that summon such behaviors. In the presence of such conformity inducing situations the adversary to demonstrate the same habits or customs.

In these eliciting situations the

adversary's behavior Is highly autocorrelated,

and shows low variance over these

situations. *

will tend

Habits are reliable reactions to aroused needs and customs are

reliable reactions In response to a stable Intention to conform. of the analyst Is to recognize the autocorrelated,

The main task

rel labia behaviors and then

determine what needs, situations or events trigger these stable responses. A second form of stable Intention occurs when a form of behavior Is constantly repeated over time regardless of particular situations. have very high base rates and will appear to be normal,

These behaviors

routine behavior.

These

behaviors may serve some functional purpose (eg., regular patrolling and reconnaissance of sensitive terrain) or may be merely a tradition serving only ceremonial purposes (May Day parades).

Some behavior may lack any recognized tradi-

tion and may be repeated merely because no alternative occurs to those who

17 *1i

perform the behavior. behavior, way,"

An analyst may create a plausible explanation

for such

but Its actual "cause" may be sImply that "it was always done that

I.e., mere behavioral

inertia.

It takes )ffort to change things,

there Is no point to fixing something that Is not broken.

and

Similarly, behaviors

that cause no harm may be tolerated and even reinforced although they produce no beref its except to keep people busy.

Behavior that has been repeated under a

wide variety of situations and circumstances with I Ittle variation In character Is likely to continue.

The Intention for such behavior may relate to identifl-

able goals or It may be simply the intention that what was done be done again. High base rate behavior Is highly predictable.

Behavior that conflicts with o'

prevents high base rate behavior will require strong motives and Intentions to overcome the inertia of the repeated behavior. Some situations bring a multitude of pressures to bear on a nation, acting to motivate behavior In a certain direction. sudden surprise attack will

all

For example, a massive

mobilize a nation's defenses,

To the degree that

the forces and pressures on the attacked nation tend to work In the same direction, the nation's behavior and Intentions are more predictable. Imagine a small nation,

unsure of Its survival,

allies, vulnerable borders and trade routes, standing armed forces,

For e>'-

with many enemies e-:'

a militaristic tradition, capable

a unified and resourceful population, a tradition of

antipathy toward the attacker,

an economic,

political,

or territorial

Interest

In the outcome of war, and a history of racial or rel igious disputes with the attacker.

Such a nation Is more predictably going to responl ml I •tarl ly to a

surprise attack than a nation with opposite characteristics and pressures.

Many

predisposing forces and pressures acting In a given direction allow a more confident prediction or estimate of Intentions then cros3-cutting and confl icting pressures.

In other words,

when a nation faces intense demands, pressures,

and f.irces to act in a certain way, action,

and lI-ttle or no opposing pressures to such

It Is I lkely to respond to the pressures.

The more pressures that act

In the same direction, the easier 'the task of forecasting the nation's Intentions. There Is a nontrivial

danger

in using a causal

pressures on a nation to estimate Its Intentions,

analysis of the converging

however.

What may appear to

the analyst to be an IrresIstable pressure might in fact pose only a temporary problem to the adversary nation. trivial

On the other hand,

what appears to be a

factor to the analyst might be an overwhelming demand that the leaders

of the adversary state cannot Ignore.

An example of the former case might be

Egypt prior io the 1973 attack on Israel.

Egypt seemingly faced severe limita-

tions of military capOs IIty which would prevent any successful

attack.

In

fact, Egypt had evolved alternative solutions to Jts military limitations (eg., Ingenious means for rapidly cutting throuqh the sand dunes along the Suez Canal wilh high pressure fire hoses) and of adjusting Its military Intentions to Its mlIitary capabilIties so as to accomplish goals.

overriding political

and diplomatic

Art example ef the latter case might be Ihe underestimation by Western

analfsts prior to 1941 of the criticality of resources to Japan and the centre! ot* of ýil supplies to Japanese Intentions.

The Western embargo on oIl

sales,

an attempt to use economIc force -o pressu.re Japan to moderate Its China pal Icy, had the effecT of Increasing Japan's militaristic ambitions (to s:ure reliable oil supplies in Southeast Asia) rather than curtailing them. Just as there are behaviors that are compel led by converging pressures, there are other behaviors that require converging preparations, ie., complex and multlfaceteJ multiple actions that lead up to and prepare for some behavior

the state Intends to act out.

Such preparations are lI ke a pitcher's wind-up,

they signal an oncoming event,

even -hough they may not reveal the exact nature

19

The en

of the pitch that Is thrown or the base the pitcher may throw to. complex of preparations may suggest one single IntegraTed assoclited with disparate,

Intention,

ire

or may be

The chronology of observations

disconnected events.

(by the Intention estimating analysts) may or may not reflect the logical order of a,:ts dictated by a single plan; instead the observational chronology may result from the mxing of actions called for by several other hand,

certain

preparations

plans and operations require that various essential

be made

In predictable

preparations and none are detected, the adversary

unconnected plans. On the

order.

If a watch

is

set for these

this is a negative Indicator; it

Is preparing that particular operation.

Is unlikely

Of couse, the adversary

may anticipate such a watch and disguise such preparations or cover them with a plausible deception.

If both predisposing pressures m

for a particular behavior are absent,

there Is

necessary preparations

little

likelihood that the

adversary.Intends to perform that behavior. Certain planned actions by an adversary require various acts be taken or processes be set in motion to carry out the Intentlon,

The attack on Pearl

Harbor for example, required that the Japanese fleet sail to Hawaii. may partially determine the courses of action open to the adversary. some point on Its eouU'e to HawaVl attack (e.g.,

if

it

These acts While at

the Japanese fleet could have aborted the

were discovered

by U.S.

reconnaissance),

particular moment at which there could be no turning back -

there was a

the air raid had +o

be launched regardless of whatever else was taking place. Some plans by an adversary may be sufficient determined by their early 4vants that the conclusions can be predicted.

The analyst observing the

unfol ding of the plan may be able to anticipate the ultimate target by noting that certain options and dlreactons have been closed off as the operation has progressed,

Indicating the Intended direction of the adversary's actions.

To the extent that the advarsary's behavior Is under the control of some agency that the analyst is able to monitor, the behavior Is-more predictable. For example,

In combat the analyst's own nation will conduct operations which

may determine the adversary's reactions.

If the analyst knows of the impending

operations against the adversary, he or she can anticipate the Impact on the The analyst's nation may be able to control the adver-

adversary's response.

sary's access to information, as for example In Britain In World War HI when all espionage agents were captured and controlled by British security forces.

Nazi

This control of Information can al low the analyst to better model the adversary's

decision process by providing direct access to some of the Information

"that gco.es Into that process. An example of how Informatlcn control assisted Intel I Igence analysts to estimate intentions occurred In the Pacific in World War II prior' to the battle Through code-breaking,

of Midway.

U.S. naval

Intelligence was able to determine

theat the Japanese Intended to launch a major attack to destroy the few remaining U.S. aircraft carriers.

The Japanese Intended to draw on the U.S. carriers by

firs? attacking a key U.S. land base.

The question unanswered by the code-

breaking was where this land at+ack was to take place:

Midway, Oahu,

the

Aleutians, the U.S. west coast were all pnssibilities with MIJway being the prime candidate.

To confirm

Intellilgence

Novay

Information thae

that the Japanese

conducted

an

Intended to attack Midway,

*Intelligence experimentQ

Midway Island was short of water.

U.S.

!taking the

This was pick.ed up by Jap-

anese Iistening posts which reported to Tokyo that the target for the forthcoming attack was having water problems.

These signals were intercepted and

read by the U.S. Navy confirmIng the suspicion that the Japanese were timlng at NMI dway.

The successful trap set by the U.S. Navy for the Japanese depended on

this confirmation by information con1'rol.



+!i

- -

w -

*..........

Analyst Lglc and Reasoning Processes It

is possible to separate Intention estimation into iirlmary steps of

estimation logic, and then determine where parts of the process might be strong or weak.

For this we divide est

t

mation Into seven different steps needed to

reach an analytic con'iuslon (see Figure 3).

These steps are somewhat arbltrary

and could be comb!ned Into fever or expanded Into more steps.

Nonetheless, they

seem characteristic of Inferenice and deductive reasoning as sludlles by cogn!tlve psychologists and seem to capture the various capabil Itles of IndivIduals to process information.

All intention estimation involves at some point The think-

Ing of Individual analysts so an appraisal of the analytic logic needed to make estimates should serve to reveal some of the nonctvious perennial

difficultles

ot Intention estImation. The mental methods people use to process Information have been labeled

"*heuristics". These are usually adequate for the reasoning we typically need to do.

Certain problems go beyond the capabilities of these heuristics,

and require more elaborate, everyday heuristics. A

less speedy,

Often It

however,

and less afficlent methods than

is dlfficult to recognize that the everyday

heuristics of thinking are inadequate and people continue to use them when they should use stronger reasoning methods. Perciving n)ata One Auch heurIstIc, fvaIlIab I Ityo data (see Figure 4). occurred,

Is used when people perceive and encide,

An analyst often decides wnether an evqnt has often

and Is thus typical, or whethar it is rare and unusual behavior.

Various characteristI cs of events influence ho, frequency Of their oCcurrence, vIvidness.

•.-----•

meeorable

they are;

the sal lence to the analyst of the e*ent,

Frequent. sel Ilnt, vIviI events are more memoreble.

the its

.'4emorabilIty In

Figure 3.

I;

II SI

-•i 1

_I•-

!23

Steps in Estimation Logic

o

Perceiving Data

o

Weighting Data

o

Characterizing Data

o

Assessing Covariations

o

Assessing Causes and Effects

o

Prediction

o

Theorizing

S

'

.,-*. _-

Figure 4.

..

..

,

...

Perceiving Data: Availability

ENCODING DATA

INFERENCE

Data Characteristics

Frequency Salience

Vividness

1

Frequency or •likelihood Memorabil

Ease of encoding or construction

24

-

Salience, vividness,

etc.

turn Influences judgements of frequency and I ikel ihood.

The events we remember

best we tend to bel leve are more likely or more frequent. we can bel ieve an event

is

frequent even

characteristic of the event that led to It

frequency

being memorable.

vivid or salient event as frequent, even when it well.

if

Note,

however,

that

were not the

We may perceive a

is not, because we remember It

Or to put It the other way, we might remember something very well because

it was sal lent or vivid, and then, because it was memorable, estimate that it was frequent or likely. reason,

The more events are available to memory,

for whatever

the more we tend to overestimate their frequency and likelihood.

Uncommon events may be perceived as common. in intention

estimation,

characteristics of events other ±hb

frequency or typicalIty seem to enhance their memorability and, analysts'

estimatt~s that these are typical behaviors.

in turn,

their inflate

International crises are

highly sallent and vivid to the Intelligence analysis community and very memor

Ile.

It Is likely that analysts tend somewhat to overestimate the I Ikel I-

hood that an adversary will shift from noncrisis to crisis behavior in periods

t

of tension. The "false alarm" rate for predictions of adversary crisis action will thus tend to be high. The proximity of one event in tIme or space to another event al lows the two to be more easily remembered as connected.

Such memorab IIty of a connectIon

may lea- analysts to perceive coincidental

events as causally related.

One

example seems to have occurred In the case of the 1969 Libyan coup and the 1973 Cold War.

The mere presence of Soviet Navy ships plus these two events

seemingly led analysts to suspect causal connections between the events and the Soviets which better evidence tends not to support. Causal tics.

links between events are among the most memorable data characterIs-

To the degree that an analyst perceives a causal

25

I Ink between an adver-

sary's behavior and U.S. remembered,

behavior,

I Ink I Ikely to be

not only Is the causal

lead the analyst

but there Is the rIsk that thIs memorability will

to overestimate the frequency with which such causal likely In the future.

Causal

Influence takes place or Is how-

links are difficult or Impossible to prove,

as is discussed but are readily perceived, even when they don't exist, Sever, further below.

So the combination of a vivid, memorable,

but spurious,

causal

link, and the availability heuristic, could lead an analyst to overestimate the likelihood of a causal relationship that, In fact, does not really exist, or is extremely rare.

Again,

false alarms about what the adversary

Is able to cause,

or Intends to cause, are likely. A second heuristic which

affects the way

"Itrepresentativeness4 (see Figure 5).

analysts perceive

data

I.s

When the analyst assesses an event,

certain characteristics and features of the event lead him to Infer the event Is of one type rather than another,

and to estimate what population of events this

particular one came from (e.g.,

is this particular behavior "hostile" or just

"unfriendly"?).

There Is nothing wrong with this logic IU the analyst gives at

least some thought to the population pool any event, regardless of Its characteristics, behavior (e.g., "unfriendly"?).

In general,

itself; asking If It

Is I lkely that

came from that particular pool of

Is the adversary's behavior more often "hostile" or

Some behavior populations are highly Improbable (e.g.,

the

launching of surprise attacks) making It unl ikely that a given event under study came from that population, no matter how strong the resemblance might be between the features of the particular event and the character of that population of events.

(That Is, many more behaviors may 12*k like preparations for surprise

attacks than actually are preparations for surprise attacks.)

Unless the

analyst considers the base rate frequency of the population of behaviors as well as considering the features of the specifIc event,

26

there Is the possIbIlIty he

Figure 5.

Perceiving Data:

Representativeness

DATA SAMPLE FEATURE 1 DATA SAMPLE FEATURE 2

MEMBERSHIP

ESTIMATED

INFERENCE

POPULATION

-,

DATA SAMPLE FEATURE 31"

27

or she will overestimate atypical and unusual events and perceive them as more common than they real ly are. Analysts who attend to the doctrinal

statements of an adversary may find a

statement or set of pronouncements that suggest a particular future intention. The specific characteristics of these statements may be very suggestive of a future action, e.g., open hostility with other countries. these particular

These features of

statements have to be balanced against the base rate of

analogous statements and the base rate of behavior. analogous statements reflects a peaceful outlook,

If the entire body of

or a cautious at-titude,

analyst should reduce the weight given to the hostile remarks. the nation being monitored has had a history of peaceful

Similarly,

the If

relations the analyst

should consider the hostile statements of representing an unusual

departure and

a possible anomaly rather than a clear-cut indicator of the future. Similarly,

analysts often make major deductions about an adversary's

tactics or strategy using military equipment construction as evidence.

That Is,

the construction of a particular weapon Is taken as evidence of a particular Intention.

Such arguments may

lose sIgnt of the other naval

or mIlItary

constructlon programs which may represent even greater emphasls or effort.

The

use of naval events, for example, as evidence of naval poI Icy or doctrine may fall victim to the same problems, that is,

using rare and unusual events to draw

conclusions about typical or general naval behavior. A fascinating example of the tenuousness of estimates of enemy Intentions extrapolated

from estimates

of

enemy

McLachlan's (1968) account of Bri•ish Naval 2

military capabil ities is

given in

Intelligence before and during World

War II.

In 1936, when the Germans were constructing their great battleship

Fl .murk,

the consensus of British diplomatic opinion was that Germany would

adhere to the Anglo-German Naval Treaty of 1936, which I Imited Germen battleship

size to 35,000 tons.

This consensus provided the basis for the design of the

British battleship KIng George V. The dimensions of the 81smark and TIr.Ltz as released by the Germans Indicated that if these ships did In fact displace only 35,000 tons as the Germans claimed, British ship.

they were of much shallower draft than the

Although the lower tonnage was doubted by Naval Intelligence

nontechnical officers, Intelligence opinion was divided.

From the I nf erred

characteristics (shallow draft and 35,000 tons) The British Naval Plans Division '-oncluded "the present delsgn of German capital

ships appears to show that

Germany Is looking towards the Baltic with Its shallow approaches more than In 4

the past" (quoted by McLachlan, more at Russia than at Britain. be roughI y 45,000 tons,

1968: 136), that Is,

the German ships were aimed

In fact, Stsmark and TlrpfLz were designed to

and the Germans released false figures to the British,

depending on British bel lef In the readiness of the Germans to honor the 1936 agreement.

The Germans read their opponent wel I; the British Director of Plans

at the time wrote, "our principal safe-guard against such an Infraction of treaty oblIgatIons I Ios In the good faith of 'the signatories" (p.

137 in McLach Ian).

Not only had Germany deceived Britain as to her capabilities, additional,

she had the

unintended benefIt of an erroneous British estimate of Germany's

naval Intentions.

Similar underestimates based on similar German deceptions

regarding submarines,

cruisers,

and battlecruisers also occurred prior to World

War II and may also have misled British estimates of Germany's willingness to engage British naval power with what the British took to be a far less capable navy than Germany In fact possessed. Similarly, U.S. un erestimates of the range and performance of the Japanese Zero and the estimate that shallow water torpedo attacks were infeasible probably Contributed to the Japanese surprise In attacking Pearl Harbor (Wohisetter, 1962: 394).

l'

29

-

- -- :

-*

Fear of the Luffwaffels strategic bombing capability, overestimated In 1938 by the British Air Staff,

Inflated British Cabinet estimates of Germany's

willingness to go to war over the Czechoslovakia crisis and contributed to Chamberlain's motivation to capitulate to Hitler, (see Bracken, Finally,

according to some historians

1977). It

is quite easy to perceive plausible lduses for almost any

event, and to detect causes that have no actual connections to events, and which may be quite rare and atypical.

Features of a data sample may lead the analyst Is Important that analysts stop to ask

to perceive a causal relationship, but it if

such relations are generally

likely or typical.

If not,

the

analyst should

lower his estimate that such unusual causal behavior Is occuring In the case under study. We ight ing Date Not only must analysts perceive data, they must weight It,

deciding which

pieces of Information are Important and which are less critical to their judgments (see Figure 6).

However,

psychologists have found that,

In general,

people are not at all accurate In Identify ng what Information they actual ly use to make judgments and decisions.

People also often use highly redundant evi-

dence as If it were completely Independent Information, two sources what Is really Just one. stable cues, which is logical,

In effect,

counting as

Cues which vary are given more weight than

but people sometimes do not notice when a vari-

able cue becomes stable, and may go on giving It I Ight weight on the mistaken assumption that It continues to vary. more ccnfIdance people feel

The more cues that are available, the

about their estimates,

even though they often faIl

to Integrate all the additional Information Into their judgments, or to make more accurate judgments with the additional

information.

People seek out those

data cues which are most I Ike the answer that is being sought.

30

To estimate a

Problems in Assigning Weight to Data

Figure 6.

I - •: -: -•• ... • • .

o

Lack of introspective accuracy

o

Reliance on redundant or highly correlated cues

o

Emphasis on variable cues

o

More cues, more confidence

o

Cue-response compatibility

o

Salience and vividness --

Case studies versus base rates

--

Neglect of nonevents

, -•

... , , o• •. -• • ••,•...... 31

: -: • .... .

?

future attack submarine threat, for example,

an analyst might seek data on

attack submarines, although data on naval aircraft or space systems might be equally Important to answer the question. given more weight than statistical

Highly sal lent and vivid data are

data, or information on nonevents.

For

example, although analysts sometimes use nonevents In their estimates, Important nonevents are often not attended or analyzed.

Categorizing nata As analysts perceive and weigh data they also categorize them Involves the act of categorizing.

That is,

-

perception

people attach a description to a

piece of Information to code It for memory and later Inferential work.

Mental

theories and conceptions aid in this process of classifying and organizing perceptions and facts. Perhaps the major difficulty people have In categorizing complex phenomena is

In trying to impose either/or

distinctions on objects or events which have

too many dimensions and too few cut and dried boundaries to be so narrowly pigeon-holed.

Psychologists have found that people typically do not rely on

either/or categorizations of complex objects or events In making judgments but Instead seem to make judgments as to whether the object is more or less prototype or schematic model of the category

In question.

lIke a

In other words,

category boundaries are fuzzy and probablistlc rather than hard and fast. The fuzzy nature of complex natural

categories and the difficulty

In

specifying all The features which give an object or event a family resemblance to a category may contribu~e Importantly to conf IIcts between Intel I igence

analysts.

For example,

t"ere are characteristics of Soviet ships that suggest

they vould be effective deterents to interventions during crises by Western aircraft carrier task groups.

These and other characteristics of thes* ships

32

give them capabilities to serve in a "blue belt defense plan" against nuclear strikes from the see on the Soviet homeland.

Simllarly,

deployments of Soviet

ships share features that relate to both the diplomatic and strategic missions. Oecidlng whether a particular event was diplomatic or strategic in character can be very difficult because of the fuzzy boundaries of these categories.

Often

analysts cani do no more than to conclude that a particular *vent was more lI ke one category and less I Ike the other but also a lI ttle like both.

Nor does It

seem to be any easier to categorize doctrinal writings, military equipment construction, operational or deployment patterns, or diplomatic events; all these complex events can be categorized as having a family resemblance to strategic or diplomatic missions, offensive or defensive objectives, cautious or expansIve tendencIes.

At best the analyst will

be abl

to make only probablIs-

tic judgments as to the categories In which an event belongs.

Oebating pigeon-

holes is far less usaful than attempting to measure the strength of family resemblances. It

Is Important, when generalizing from a sample of data, to be conscious

that certain data sampling methods bias the Information in predicatble ways. One such bIas Involves rolis. I.e.,

For example,

the role of deterrer in a cr itIs,

the party that acts to deter another party,

bias.

Namely,

has a built-in role conferred

If the adversary does nothIng, the deterrer can credIlIy claIm

success In detarrIng, even though the deterrer actually may have Jone ý!'title and the lack of action by the deterred was due to other factors (e.g., te lackeod any Incentive to act).

The

Weterrer Is

adversary

a little elks the man who

scatters corn flakes around himself to kepthe engal aters point out *hat there are no tigers In +he nelghboihood,

way.

Whenyou

he beams at you and

.akes cradit for his &auing deterrent ofoers. A seco'd tfpe of sampl Ing bias occurs when aft analyst a*tends a part but

33

T not all of a data confIguratIon example,

(see Figure 7).

the entire lata configuration is shown.

In panel A of the hypothet'icat Various biases can be seen In

panels B, C, and 0. When the whole pattern Is considered all one would want to conclude Is that SovIet naval events of all types are twIce as frequent latter time pericd. J.

In the

Biased sampling, however, might lead to the conclusions

that Soviet diplomatic naval activity has doubled, or that strategic forward deployment has doubled, or that the SovIets are now twice as likely to engage In strategic as opposed to diplomatic forward deployments. Real examples of such biased sampling can be found In naval analysis. example,

McGwire (1976:

For

166) argued ttkat the Soviets surveyed the Indian Ocean

from 1967 to 1969, making many diplomatic port visits to scout out facilities, and then visited very few ports thereafter, having found the facilities they needed.

Petertn (in Dismukes and McConnell,

1979: 91-92) disagreed,

writing:

While It Is true that a drop In diplomatic vistis to Indian Ocean ports was registered In 1970, 1' is na= true that *very fewvI visits have been made elsewhere In the region since then. Between 1970 and 1971, for example, no fewer than 30 diplomatic visits were made to Indian Ocean countries other than SomalIa and South Yemen. In comparison, only 28 were made to Mediterranean ports during the same perimo. The total data configuration Ii shown In the figure (see Figure 8).

There

were nearly twice as many Indian Ocean visits as Me•iterranean ones, but In the laiter period the ratio of visits Is not 2 to I but 1.4 to 1. While vltsl year Increased In the M1a terranean over the two periods from 2.3 to 4, Indian Ocean they decreased from 10 per year to 5.6.

per in the

Petersen's comparison of

30 to 28 visits is biased and misrepresents the far higher level of overall vislting in the Indian Ocean and the sharp decl ;e

of visiting in tha+ ocean In

The latter time period to a level comparaole to That In the Mediterranea.n

34

Figure 7. Hypothetical Example of Biased Sampling

A. Soviet Naval Events Diplomatic Forward

Strategic Forward

1954-1967

15

30

45

1964-1974

30

60

90

45

90

135

Total 1954-1974

B.

Column Bias:

(a)

Total

Diplomatic Forward Perspective

(b) Strategic Forward Perspective

1954-1963 1964-1974

C.

Row Bias:

Strategic

Forward

Forward

15 30

30 60

New Soviet Navy Perspective

1964-1974

D.

Diplomatic

Total Bias:

Diplomatic Forward

Strategic Forward

30

60

Overall Perspective Diplomatic Forward

Total 1954-1974

Strategic Forward

45

35

Figure 8. Number (and Percentage of Diplomatic Port Visits in the Indian Ocean and Mediterranean by the Soviet Navy 1967-1969, 1970-1976*

1967-1969 1970-1976 Total 19671976

I

Indian Ocean

Mediterranean

Total

30 (43%) 39 (57%)

7 (20%) 28 (80%)

37 67

35 (100%)

69 (100%)

104

-*

Data from Petersen, Table 3.2 (p. 92 in Dismukes and McConnell, 1979.)

36

2'ata

Coverlatlons Analysts must be able to discern relationships between sets of data to Infer that one set of evor,-s covarles with another set.

For example,

military

analysts must assess the covarlation between an adversary's strategic needs and military activities, between political commitments and crisis behavior, between capabilities and actions, etc., to understand and predict military Intentions. People have difficulty perceiving unexpected covarlations and tend to perceive expected covarlations even when there are none.

That Is, people are

far too ready to detect theory-driven covarlations in the data they process and I/

unable to see unexpected data-driven covarlations unless they are overwhelmingly strong.

Psychologists have found that people underestimate covarlations,

I.e.,

strong correlations (e.g., r = .6 to .8) are perceived as weak relations, and

j

near perfect covarlations (r = .8 to 1.0) are perceived as merely strong relations.

Unexpected relationships, even when near perfect, tend to be missed.

Unless an analyst expects to find a covariation relationship, he or she will tend not to notice one unless It

Is very strong.

relations may go unnoticed, or underestimated.

Potential

Consequently,

strong

Indicators of future

activity may be overlooked and important predictable patterns may be neglected. On the other hand,

psychologists fInd that people will

perceive an

expecteo, theory-driven covarlation even when none exists In the data set. seems to happen Is that positive occurrences are noted, the two expected events do, In fact, occur together. not the other, event occur are neglected.

What

I.e., those cases when

Those cases where one, but

The analyst thus amasses a convincing

list of confirmations of the relatlonship but has Ignored all the dIsconfIrmIng cases. Such difficulties In assessing covarlations can be found in naval analysis (see Figure 9).

For example, one analyst argued that Soviet diplomatic naval

37

Problems in

Figure 9.

o

o

Aszassing Covariations

Theory-driven correiaticns are perceived, in the data.

even if

absent

Theory:

The expansion of Soviet diplomatic visits in the Indian Ocea, 1967-1976.

Data:

Correlation between year and number -. 38. of visits:

Data-driven correlations are not perceived

if

absent from

theory. Theory:

The intensity of diplomatic visits reflects the prominance of political concerns in the Indian Ocean squadron's mission structure.

Data:

Correlation between ship days and diplomatic visits: -. 65. Correlation between ship days and lengths of visit: -. 83.

o

Illusory Correlations e.g.,

naval presence and coup occurrences.

38

visits in the Indian Ocean expanded In the years 1967 to 1976, reflecting Soviet interest in tf.!t area. as the years go by.

This implies a positive relation;

In fact the relation Is negative:

as the years went by In this period.

more visits per year

there were less visits

The data suggest there was a contraction,

not an expansion, of Soviet diplomatic visits In the Indian Ocean. hand,

unexpected correlations can go unnoted.

Soviet Naval

On the other

One analyst expected that as the

Squadron spent more ship days In the Indlan Ocean,

of Soviet political concerns would Intensify

the prominance

diplomatic visits.

The data

reflected just the opposite; the more ship days the Soviets spent In ,1he Indian Ocean,

the fewer the number and the shorter the duration of diplomatic visits.

These unexpected

(and unnoted) negative relations were quite strong.

when a theory suggests a relatlonshfp should exist, coincidental may be taken as confirming evidence.

Finally,

co-occurrences

On several occasions the presence In the

area of Soviet ships has been Interpreted as related to the occurrence of coups, or crises, although there was little

evidence to substantiate any connection

between the events and the Sovietfs presence. Cause and Eff2G Covariations are Important In their own right, but also because they are crucial data for Inferring cause and effect relationships.

The analysis of

causes Is one of the Intelligence analyst's major objectives; having causal understanding of an adversary enables you to predict his future actions. Determining causes Is one of the most difficult of Intellectual and Information processing operations, however,

and one for which a myriad of analytic aids are

not just useful but often necessary.

Our Intuitive,

judgments about cause and effect are often In error.

unaided reasoning and

Even scientists sometimes

find nonexistent causes for events. Psychologists fInd that there Is a strong tendency to perceive as having a

39

1z'

causal role anything on which we focus our attention when considering cause and effect (see Figure 10).

Futher,

if

an actor is perceived to have a certain

Intention, and arn even-r occurs bringing about the desired outcome, there Is a strong tendency to view the actor

as having caused the event, although the event

may have occurred for other reasons. are observing to their Intentions,

We attribute the behaviors of actors we

dispositions,

and motives and underestimate

the degree to which their behavior is controlled by the environment,

the situa-

tion, or the contoxt they are in. To the degree an ac.tor's behavior has Important consequences and the actor can foresee the consequences of his actions, behavior to a profound motive. effect:

That Is,

if the effect is profound,

profound.

the more we attribute the actor's

the cause is perceived to match the

the motive, or cause, must also be equally

This is ai msnifestation of the representativeness heuristic, the

tendency to assume thait features of the cause should be representative of the feature:s of the effect.& The problem is that rather trivial causes and motives can i•ead to very profoL.nd consequences. If an actor is oer°celved +o foresee his actiomis' consequences, tends to attribute the actions to Intention.

For example, the Soviet Union's

war f I htIng outlook ImplIes foresIght; the anticipation of general There Is a tendency to assume that,

the observer

nuclear war.

If the Soviets are able to plan and prepare

for such a war, they may Intend to wage such a war to win their own ends. Soviets may thus be seen as building a nuclear blackmail It

capability.

The

Of course,

In also possible that the Soviets have no confidence In their own deterrent

capability,

and prepare for nuclear war,

not optimistically for blackmail,

pessimistically out of a tear that they cannot forestal;

but

It.

In "magical thinking", the features of effects are used to guide the search for causes

--

causes a-e sought whicn resemble effects.

40

Perhaps the worst

Figure 10.

Problems in Assessing Causes and Effects

o

Dispositional vs. situational attributions e.g., explanations of forward deployment

o

Profound motive fallacy

o

Foreseeability implies causality

o

magical thinking --

causes resemble effects

e.g., capabilities cause intentions, can do will do

--

o

Minimal causation

a

Causal. hydraulics e.g., desire for military superiority precludes desir-e or detente

o

Parsimony and "Indiscriminant Pluralism"

I4

example of magical thinking is adversary's

intentions

capabilities (effects).

(causes)

when the analyst lets himself believe the follow

In particular,

from

or

resemble

their

military

there is the danger of calculating the

worst the adversary cguld do and then reasoning that that is what the adversary will do.

Alternatively,

the analyst may be tempted to calculate the limits of

the adversary's capability and then estimate that the adversary would not undertake missions that exceed those apparent limits.

"Minimal causation"

refers to the tendency to accept the f.Lrst reasonable cause that fits the as thezIj4 cause.

1ta

After people find a pllausible cause to explain an effect,

they seem to stop searching and the hypothesis that multiple causes contribute to an effect is rarely tested.

"*Causal hydraulics* are a colorful

way of describing the tendency to

bellev,: +hat causes compete and compensate for each other In producing effects. That Is,

an analyst might believe that If strong political forces are producing

some behavior, tnen military or naval factors cannot also be contributing at the same time.

The notion that bureaucratic politics shape an adversury's behavior

and that competing agencies and Interests are contending for Influence Is especially subject to causal hydraulic reasoning.

The danger Is that necessary

causes may be mislabeled as sufficient causes. A finaI problem In causal reasoning Is the tendency to seek a plausible cause for every behavior.

Ilndlscrmllnant

plural Ism"

This can easi y lead to what one historian termed -

an attempt to propose a cause for everything.

The

problem with efforts to explaIn everythIng is that one ends up wIth too many explanations and no ability to predict.

There is no way of knowing which

explanation to use for the future situation. parsimony,

f

This Is why scilntlsts seek

the fewest possible causes suffIcient to explain and predict.

The

smaller the number of causal explanations that can account for events,

the

better the chance these causes will predict the future.

Predicting and forecasting are central tasks in

Intelligence analysis.

Analysts, however, have been charged with being unable to accurately predict either the capabilities or intentions of adversaries.

Some recent prediction

failures are shown in the figure (see Figure 11). Many explanations of failures of

Intelligence predictions seek the causes

In unique characteristics of Intelligence estimation.

These explanations are

made less compel ling by the fact tha forecasting and predicting are not done with much accuracy In other fields.

Recent reviews by Ascher (1978) and by

Hogarth and Makridakis (1979) of population,

economics,

energy,

transportation,

technology, and business forecasts found that prediction In these areas Is rarely accurate. l.•

general ones,

This suggests that problems of forecasting and prediction are

not limited to Intelligence analysis.

People need to perceive their world as orderly and predictable.

The more

people believe they control events or that events are controllable (even if such control

Is Illusory),

the more predictible the world seems to them.

This illu-

sion cf control can lead to predictions made with great con,;Idence of events that are In fact controlled by chance. more predictable than It

Forecasters may perceive the future as

Is simply because of their efforts to predict it.

People see pattern and order where none exists because of their need for an orderly world.

There are extremely strong perceptual

the environment and make sense out of It

tendencies to structure

In order to organize perceptions.

Pople rarely consider the possibility that the environment may have random or probabil Istlc elements,

Instead they see events as determined and fixed or

ordered by regular mechanistic (not probablilstlc) processes,

43

Figure 11.

o

Prediction Problems

Prediction Failures - Bombers - ICBMs - IRBMs - Launchers -

MIRVs

-

Defense Expenditures Surprise Attacks Crises Peace Offensives

-

o

General Causes -

Illusion of Control

-

Illusion of Order Faulty Assumptions Unreliability of Judgment Vice

-

Statistical Relations Illusions of validity and overconfidence

-

-

Inappropriate Techniques Theory-driven overprediction

44

r Forecasters often reach different conclusions which all reflect a similar biased inaccuracy.

Ascher (1978) concluded that forecasters making estimates in

the same time period all tend to share the same key assumptions, and to the degree these are wrong, tend to make the same mistakes.

The analysts' key or

core assumptions may be much less carefully scrutinized than data or methods, but bad core assumptions

may produce greater error than

flawed

data or

I nappropr I ate methods.

For many mediumextrapolations

and short-term forecasting tasks, simple statistical trend

are often more accurate

than are analytic judgments,

or

theoretically elaborate or methodologically elegant forecasting methods.

People

have very high confidence In the superior accuracy of their judgments,

a level

of confidence unwarranted

by the demonstrated

relative to simple statlstlcal methods.

inaccuracy

of their predictions

People are sometimes most confident

when they are least accurate. Often forecasters employ highly inappropriate techniques which virtual ly assure an inaccurate prediction. Delphi.

One very overworked and abused technique

Is

The famous economist, Paul Samuelson, noted that:

the greatest error in forecasting Is not realizing how important are the probablI Itles of events other than those everyone Is agreeing upon. To the extent that Samuelson Is right (and there are several reasons why he probably

Is) then a technique like Delphi

Is doomed to Inaccurate

predictions

because it focuses the experts' attention upon the probabilities on which everyone has agreed. and the unexpected,

Delphi

systematically seeks out the extreme predictions

low probabIII ty conjectures,

leavIng only the middle range,

concensus probabIlItIes most of the experts had already thought about. extremes that Delphi eliminates,

however,

45

The

are the most likely to predict

surprises and unexpected events.

Delphi Is appropriate only if surprises or

unexpected events are not going to occur. degree of orderliness,

But If we can be assured of that

we do not need Delphi; simple statistical extrapolations

will do better and cost less.

Theorc The use o; theory, the last analytic step, Is much to be commended In any prediction or forecasting effort.

There Is the chance, however,

will suggest more order and predlctabillty Ing analysts to over-predict.

That is,

that the theory

in The data than really exist,

lead-

the analyst with a theory may make more

extreme predIctIons wIth greater confidence than the data and the analIyst's prediction record would warrant (see Figure 12). using theories is to make accurate predictions.

One of the main reasons for The analyst must take care not

to use theory wishfully, and to keep o~rdtctIng distinct from theory teting. Creating and modifying theories of the adversary should be one of the major goals of the Intelligence analyst.

Whether thel acknowledge them or not,

analysts are always using something like a theory as they proceed through the previous six estimation steps.

The danger of using theory Is that It can so

easily and pervasively color the other data processing steps, tions of data, covarlations,

leading to percep-

and cause and effect that are not accurate or

valid. Philosophers of sclence argue that the best procedure for refInIng and testing theory is to try to disprove It

In whole or part.

ti

general,

however,

people (including scientists) tend to test theorles by the tar less productive and less valid method of accumulating confirming evidence. vIves v repeated 4,Tacks

A theory that sur-

Is IIkely to be more val Id than one that has masses of

confirming data to support it but has never been subjec t ed to a olsconfirmatlon

46 V ,

6


•---

-:•< -

'

-

i ..-

.m--

= .

.":•• ..

- "-

:.:



'





--

:.

...

"

--- -

" ...

.

...

..

"-•----"